Rethinking electricity consumption and economic growth nexus in Turkey : environmental pros and cons

Author’s accepted manuscript (postprint)

Rethinking electricity consumption and economic growth nexus in Turkey: environmental pros and cons

Etokakpan, M. U., Osundina, O. A., Bekun, F. V. & Sarkodie, S. A.

Published in: Environmental Science and Pollution Research DOI: 10.1007/s11356-020-09612-4

Available online: 08 Jul 2020 Citation:

Etokakpan, M. U., Osundina, O. A., Bekun, F. V. & Sarkodie, S. A. (2020). Rethinking electricity consumption and economic growth nexus in Turkey: environmental pros and cons.

Environmental Science and Pollution Research, 27(31), 39222-39240. doi: 10.1007/s11356- 020-09612-4

This is a post-peer-review, pre-copyedit version of an article published in Environmental Science and Pollution Research. The final authenticated version is available online at:

https://link.springer.com/article/10.1007/s11356-020-09612-4

1

Rethinking Electricity Consumption and Economic Growth Nexus in Turkey:

1

Environmental Pros and Cons

2 3

Mfonobong Udom ETOKAKPAN 4

Department of Economics, Famagusta, 5

Eastern Mediterranean University, North Cyprus, via Mersin 10, Turkey 6

&

7

Economics Department, Babcock University, Ogun State, Nigeria.

8

Email: [email protected] 9

10

Olawumi Abeni OSUNDINA 11

Economics Department, Babcock University, Ogun State, Nigeria.

12

Email: [email protected] 13

14

Festus Victor BEKUN^{a, b} 15

aFaculty of Economics Administrative and Social sciences, 16

Istanbul Gelisim University, Istanbul, Turkey 17

&

18

bDepartment of Accounting, Analysis, and Audit 19

School of Economics and Management 20

South Ural State University, 76, Lenin Aven., 21

Chelyabinsk, Russia 454080.

22

E-mail: [email protected] 23

Email: [email protected] 24

25

Samuel Asumadu SARKODIE¹ 26

Nord University Business School (HHN). Post Box 1490, 8049 Bodø, Norway. Email:

27

[email protected] 28

29

1 Corresponding author: Samuel Asumadu SARKODIE: Email: [email protected]

2 Abstract

30

The critical role of electricity consumption in influencing and reshaping the economic and 31

environmental landscape of the global economy cannot be underestimated. Electricity is the most 32

beneficial and commonly transformed energy source, however, the strength, weakness, opportunities 33

and threat of its consumption requires scientific scrutiny. This study investigates electricity-led growth 34

hypothesis vis-à-vis its impact on the economic growth and the environmental quality of Turkey. The 35

annual time series data set from 1970 to 2014 were employed in the analysis with a battery of unit root 36

and stationary tests. The equilibrium relationship in the study is explored using Maki and Bayer &

37

Hanck combined cointegration tests under multiple structural breaks along with the Pesaran’s ARDL 38

bounds test procedure for a robust check. The study confirms the existence of a cointegration 39

relationship between electricity consumption, economic growth, capital, labour and ecological 40

footprint. To detect the direction of causal relations, the VECM Granger causality test is employed.

41

The causality analysis provides empirical evidence that supports the electricity-induced growth 42

hypothesis in Turkey. This implies that embarking on conservative energy-efficient policies will slow 43

down Turkey’s economic growth. Thus, precautionary measures that ensure adequate policy on energy 44

mix to guarantee availability and accessibility to modern electricity will sustain economic growth and 45

improve environmental sustainability.

46

Keywords: energy conservation, energy-efficient, environmental pollution, cointegration analysis, 47

Turkey.

48 49

3 1. Introduction

50

Following the seminal study on the US economy, the relationship between energy (electricity) 51

consumption and economic growth has received much attention in the energy economics literature 52

(Kraft and Kraft, 1978). Subsequent studies include Owusu and Asumadu-Sarkodie (2016), Alola and 53

Alola (2018), Emir and Bekun (2019), Sarkodie and Adams (2018), Akadiri et al. (2019), Bekun et al.

54

(2019a, 2019b), and Shahbaz et al. (2019). However, the documented studies report divergent 55

empirical findings, as no consensus has been reached on the nature of the relationship. According to 56

the recent statistical report by the US Energy Information Administration (EIA, 2018), there exists a 57

strong correlation between national energy consumption and economic growth. There exists a positive 58

trend between electricity (energy) consumption and economic growth (see Figure 1 in the appendix).

59

This position is further strengthened by the empirical findings of Mohiuddin et al. (2016).

60

The pertinent role of electricity consumption in the transformation of economies—whether 61

developing, emerging or developed socioeconomic landscape—has been proven in the empirical 62

literature. Electricity consumption is an integral part of a typical long-term economic growth process 63

of global economies. Unfortunately, data from the global energy market reveal that the world currently 64

experiences an energy shortage, given the global energy demand (EIA, 2018).

65

There exist a large body of theoretical studies on economic growth, bulk leverage on the well-known 66

Solow growth model (SGM). The Solow growth model depicts a substantial level of labour and capital 67

accumulation with the right level of technology known as the “Slow residual”, which explains 68

economic growth. Though technological development is outside the scope of the Solow model, the 69

endogenous growth model emphasizes the perspective of ensuring and enhancing economic growth.

70

This is possible by maximizing profit using technological progress in making a sound investment 71

decision that increases output overtime. Where deliberate effort by the economic agents are targeted 72

4

at market incentives through certain reactions, such tool or variable used is endogenous (Aghion and 73

Howitt, 2008). While the Solow growth model describes technology as physical capital, the 74

endogenous model stresses the concept of learning by doing and human capital. This duo augments 75

the marginal product of capital. This link shows the relationship between electricity consumption and 76

economic growth. The influence of this relationship does have a spillover effect within and without 77

an economy. Over the years, the conventional Solow growth model has been augmented with other 78

variables like education, tourism, population and other demographic indicators (Soytas and Sari, 2009).

79

Recently, the ecological footprint has been introduced into models as a proxy for the environment 80

(Dogan et al. 2019). This study includes ecological footprint, a composite variable, as a control variable 81

in the econometric modelling to account for environmental quality. The motivation for the inclusion 82

of ecological footprint follows several studies in the energy economics literature that utilized carbon 83

dioxide emissions (CO2) as a measure for environmental sustainability. Where there are high levels of 84

CO2 emissions, the environment suffers a negative impact from such action through pollution of all 85

sorts. CO2 is a proxy that enjoys massive recognition cannot completely capture the quality of natural 86

habitat. On the contrary, the ecological footprint captures the quality of various natural ecosystem 87

necessary to support the economy. The composite nature of the ecological footprint motivates and 88

justifies our rationale for using as a proxy variable for measuring the extent of environmental 89

degradation. Few studies have used the ecological footprint in the energy-environment and income 90

nexus literature (Katircioglu et al. 2018; Ozturk et al. 2016). Hence, the inclusion of the ecological 91

footprint is expected to add value to the existing literature in the area where samples of electricity 92

consumption and environmental proxies are involved. Contrary to previous attempt (Ghali & El- 93

Sakka, 2004; Soytas & Sari, 2009; Solarin, 2011), our study is the first to augment the electricity-led 94

growth literature by incorporating capital and labour as a case study in Turkey.

95

5

Given the mentioned arguments, this study contributes to the existing literature by analyzing the 96

relationship between socioeconomic, energy and environmental outcomes for Turkey using 97

multivariate modelling framework. We further augment for the first time the EKC hypothesis using 98

capital, labour, electricity consumption and real output for Turkey with ecological footprint adopted 99

as a proxy for environmental degradation in the energy economics literature. Using ecological 100

footprint as a measure of environmental degradation is a much broader measure compared to CO2

101

emissions. The ecological footprint incorporates among others, carbon footprint, water resources, 102

marine ecosystem footprint, grazing holding capacity and forestry (Global Footprint Network, 2018).

103

All these are unit of various natural areas needed to support an economy. Thus, the use of ecological 104

footprint is a useful indicator to measure environmental quality. The incorporation of several 105

important inputs ensures that the problem of omitted variable bias is controlled, given the level of 106

connectedness among the variables (see Kayhan et al., 2010; Shahbaz & Feridun, 2012; Tamba et al., 107

2017). The policy implication of this individual-country-based study comes with high research value 108

as opposed to panel-based studies across countries. We re-examine the SGM with the integration of 109

energy consumption as a key driver of economic growth in Turkey. This, in essence, improves the 110

existing bulk of studies on the theme under consideration by extending the scope towards an 111

interesting environmental dimension which is lacking in previous studies. Our methodological 112

innovation through the adoption of up-to-date econometric procedures enhances the precision of 113

estimates derived. Previously conducted studies on the Turkish economy mostly suffer from 114

specification bias given their bi-variate nature (see Aslan (2014) and Nazlioglu et al. (2014)). As such, 115

we fear estimates and policy recommendations emanating from such studies are unreliable.

116 117

6 2. Review of Literature

118

The pioneering work on the nexus between GNP and income (Kraft and Kraft, 1978) has birthed 119

many other studies in the energy economics literature such as Cowan et al. (2014), Farhani et al. (2014), 120

Salahuddin et al. (2015), and Bento and Moutinho (2016). Other examples include the study of Ozturk 121

and Acaravci (2011) on 11 countries in the Middle East and North Africa (MENA) region. The authors 122

investigated the electricity consumption-economic growth relationship using the Autoregressive 123

Distributed Lag (ARDL) model for the period 1971 - 2006. Their findings provided no evidence in 124

support of a significant relationship. A similar study conducted with the aid of the vector 125

autoregressive method on the Ghanaian economy by Twerefou et al. (2007) found that economic 126

growth Granger causes the consumption of both electricity and petroleum products.

127

In literature, the relationship that exists between electricity consumption and economic output is 128

classified into four categories, namely: Feedback, Growth, Conservative and Neutrality hypotheses.

129

The feedback hypothesis underlines a mutual response between electricity consumption and economic 130

growth. This is identified through a bidirectional causal relationship (Lee et al., 2008; Tang & Tan, 131

2013). The growth hypothesis posits that there is a positive monotonic relationship between electricity 132

consumption and economic growth. This scenario suggests that electricity consumption drives 133

economic growth (see Ghali & El-Sakka, 2004; Damette & Seghir, 2013). The conservative hypothesis 134

assumes a unidirectional causality from economic growth to electricity consumption. This hypothesis 135

suggests that shuffling of energy policies translate into little or no positive growth effects (Jamil &

136

Ahmad, 2010; Baranzini et al., 2013). The neutrality hypothesis postulates no causal interactions 137

between economic growth and electricity consumption. This implies that economic growth is not 138

dependent on either expansionary or conservative energy policies, particularly those targeted at 139

7

electricity consumption, as they will have no significant impact on economic output (Soytas & Sari, 140

2006; Halicioglu, 2009).

141

It is important to note that there is no unanimity in the electricity consumption-economic output 142

nexus literature as contradictory results have been reported overtime for an array of countries. For 143

instance, Yang (2000), Jumbe (2004), Yoo (2005), Tang (2008), Odhiambo (2009), Sami (2011), and 144

Shahbaz et al. (2011) report feedback causality between electricity consumption and economic growth.

145

Studies by Chang et al. (2001), Shiu and Lam (2004), Altinay and Karagol (2005), Böhm (2008), Akinlo 146

(2009), and Dlamini et al. (2015) represent instances where causality runs from electricity consumption 147

to economic growth. Ghosh (2002), Narayan and Smyth (2005), Yoo and Kim (2006), Halicioglu 148

(2007), Jamil and Ahmad (2010), Adebola et al. (2011), and Cowan et al. (2014) instead detect causal 149

relations from economic growth to electricity consumption. No causal relationship between electricity 150

consumption and economic growth has been reported by Soytas and Sari (2003), Payne (2009), Balcilar 151

et al. (2010), and Akpan and Akpan (2012). For instance, in the recent study conducted by Balcilar et 152

al.,(2019) that explored the energy growth and environment nexus for the case of turkey via the 153

adoption of Maki cointegration technique for equilibrium relationship among the interest variables.

154

The study found empirical support for the conservative hypothesis. Thus, informing policymakers 155

that embarking on energy conservative policy does not have a deteriorating impact on the Pakistan 156

economy. Conversely, the study of Bekun and Agboola (2019) joins the strands of studies that support 157

the energy (electricity) led growth hypothesis in Nigeria. This position is strengthened by the study of 158

Samu et al. (2019), for the case of Zimbabwe with an energy-dependent economy. Thus, measure(s) 159

to apply and implement energy conservative approach will hurt such economy. This is insightful and 160

informative to policymakers for proper and decisive policy formulation and implementation. A 161

detailed summary of studies on the theme over the last couple of decades is presented in Table 1.

162

8

Table 1: Summary of electricity consumption and economic growth nexus literature 163

Author(s) Time Study Area Method Causality Direction Hypothesis Ghosh (2002) 1950 -

1997

India Engle-Granger Causality test

Y ⇒ EC Conservative

Sarwar et al.

(2017)

1960 - 2014

210 countries

PECM Granger causality test

EC ⇔ Y, OP ⇔ Y, GFCF ⇔ Y

Feedback

Narayan and Smyth (2005)

1966 - 1999

Australia Cointegration Granger Causality Test

Y⇒ EC, E ⇒ EC Conservative

Dlamini et al.

(2015)

1971 - 2009

South Africa

Bootstrap rolling- window Approach

EC ⇒ Y for two sub-periods

Growth

Altınay and Karagol (2005)

1950 - 2000

Turkey Dolada and Lütkepohl (1996) Causality Test

EC ⇒ Y Growth

Cowan et al.

(2014)

1990 - 2010

BRICS countries

Bootstrap panel causality test

EC ≠ Y, EC ≠ CO2, CO2 ⇒ Y for Brazil;

EC ⇔ Y, Y⇒ EC, EC ≠ CO2, EC⇎

CO2 and CO2 ≠ Y for Russia; EC ≠ Y, EC ⇒ CO2 and

Neutrality and Growth

9

CO2 ≠ Y for India;

EC ≠ Y, EC ≠ CO2 and CO2 ≠ Y for China; and Y⇒ CO2 for South Africa Mozumder and

Marathe (2007)

1971 - 1999

Bangladesh Johansen Cointegration Test and Granger Causality Test based on VECM

Y⇒ EC Conservative

Nazlioglu et al.

(2014)

1967 - 2007

Turkey ARDL model, Linear and Non-Linear Granger Causality Test

EC ⇔ Y for linear causality test, no non-linear causality between EC and Y

Growth

Samu et al, 2019 1971-2014 Zimbabwe Zivot-Andrews, Maki

Cointegration test, Toda- Yamamoto causality test

EC⇒ Y Growth

10 Narayan and

Smyth (2009)

1974 - 2002

Middle Eastern Countries

Bootstrap Causality Approach

EC ⇔ Y Feedback

Solarin and Shahbaz (2013)

1971 – 2009

Angola ARDL Bounds Test and the VECM Granger causality test

EC ⇔ Y, U⇔ EC for the short-run;

EC ⇔ Y, U ⇒ Y and U ⇒ EC for the long-run

Feedback, Growth, Conservative

Balcilar et al.

(2010)

1960 – 2006

G-7 Countries

Bootstrap Granger non- causality test

EC ⇒ GDP for only Canada, there is no causal links between energy consumption and economic growth for the other countries

Growth, Neutrality

Akpan and Akpan (2012)

1970 - 2008

Nigeria Multivariate VECM

Y ⇒ CE, EC ≠ Y Conservative and Neutrality Shahbaz et al.

(2011)

1971 - 2009

Portugal VECM Granger causality test

Y ⇒ EC, EC ⇔ E and E⇔ Y for the short-run; Y⇔ EC, E⇔ EC and Y⇔ E for the long-run

Conservative, Feedback, Feedback, Feedback and Feedback

11 Shahbaz and

Lean (2012)

1972 - 2009

Pakistan ARDL model and Granger causality tests

EC ⇔ Y Feedback

Shahbaz and Feridun (2012)

1971 - 2008

Pakistan ARDL Bounds Test

Y⇒ EC Conservative

Soytas and Sari (2003)

1965 - 1994

Poland Cointegration and Error Correction Model

Y ≠ EC Neutrality

Mutascu et al.

(2011)

1980 - 2008

Romania Bound Test (Toda Yamamoto)

EC ⇔ Y Feedback

Chontanawat et al. (2006)

1971 - 2000

Czech Republic

Granger causality

EC ⇒ Y Growth

Narayan and Prasad (2008)

1960 -- 2002

Hungary Granger Causality

Y⇒ EC Conservative

Ozturk and Acaravci (2009)

1990 - 2006

European and Eurasian countries

Pedroni Cointegration

EC ≠ Y Neutrality

Erdal et al. (2008) 1970 - 2006

Turkey Johansen Cointegration and Granger causality

EC ⇔ Y Feedback

12 Halicioglu (2007) 1968 -

2005

Turkey ARDL,

Granger Causality

Y⇒ EC Conservative

Böhm (2008) 1960 – 2002

Slovak Republic

Granger Causality

EC ⇒ Y Growth

Yoo (2005) 1971 - 2002

Indonesia, Thailand, Malaysia and Singapore

Engle-Granger;

Granger Causality;

Johansen- Juselius

&Hsiao’s causality-VAR

Y⇒ EC, Y⇒EC, EC

⇔ Y, EC ⇔ Y

Conservative, Feedback

Notes: The symbols ‘’ ⇒, ⇔,≠’ indicate unidirectional, bidirectional causality and neutrality hypothesis, respectively. Where 164

EC is electricity consumption, FD is financial development, U is urbanization, E is employment, EI is energy intensity.

165 166

3. Methodological Construct 167

3.1 Data 168

This study explores the long-run and short-run relationship between energy consumption in our case, 169

electricity consumption and economic growth (RGDP), capital (K) and labour (L) for the case of 170

Turkey. The data for electricity consumption and real economic output were retrieved from the World 171

13

Bank database² while data for ecological footprint measured in global hectares (gha) were retrieved 172

from Global Footprint Network³. The annual data used for the econometric analysis spans 1961-2014.

173

The data description, units of measurements and sources are presented in Table 2. The variables 174

include ecological footprint (EFP) as a proxy for environmental quality, real gross domestic product 175

(RGDP) measured in constant 2010 USD, and electricity consumption measured in kWh/hr per 176

capita. Likewise, capital is measured with gross fixed capital formation constant 2010$. Labour is a 177

measure of the total labour force. This study is distinct from previous studies in terms of choice of 178

data selection. The motivation for the data choice is drawn from United Nations sustainable 179

development Goals (UNSDG 7, 8, 9 and 13). Goal 7 outlines the pivotal role of access energy use to 180

sustainable economic growth. The contribution of goal 8 is informed by improved labour productivity 181

and access to financial services (SDG 8). The advancement in Labour/Gross capital formation 182

alongside labour productivity and manufacturing output relies on investment, which in turn build 183

infrastructure and by extension spur industrial share of economic development (SDG 9). The quest 184

to mitigate the menace of global warming triggered by Greenhouse gas emissions (CO2) motivate the 185

efficient use of energy sources and its related services (SDG13).

186 187 188 189 190 191

2 Available at https://data.worldbank.org/

3 Available at https://www.footprintnetwork.org/our-work/ecological-footprint/. Note: The data span for this study span from 1990-2014 informed based on data availability especially the proxy for labour from the WDI indicators

14

Table 2: Description of data and unit of measurement 192

Source: Authors’ compilation using data from the World Bank database (WDI) and the Global 193

Footprint Network (GFN).

194 195

The empirical route of this study follows after a brief descriptive statistics comprising of mean, 196

standard deviation, maximum, minimum and correlation analysis. The path proceeds in four steps (a) 197

Investigation of unit root test properties via conventional unit root test of Augmented dickey fuller 198

(ADF), Philips Perron (PP), Elliott, Rothenberg & Stock (ERS), Dickey-Fuller generalized least 199

squares (DF-GLS) and stationarity test of Kwiatkowski, Phillips, Schmidt & Shin, (KPSS). In the case 200

of a possible structural break, the Clemente-Montanes-Reyes structural break detrend test and Zivot- 201

Andrews (ZA) are utilized to know the asymptotic properties of the investigated series. To ascertain 202

the maximum order of integration and avoid the error of working with variables integrated with ~I(2) 203

as outlined by Moutinho et al. (2018). (b) Examining the long-run equilibrium (cointegration) 204

properties of the variables under review with estimators that accommodate for possible structural 205

breaks. (c) The exploration of the long-run magnitude in terms of coefficients among the investigated 206

variables. (d) Finally, the detection of direction of causality flow among the series via the VECM- 207

Granger causality test approach. The vector error correction (VECM) model approach is the most 208

Series Name Unit of measurement Source

Real Gross domestic product (RGDP) Constant 2010 $ USD WDI

Electricity consumption (EC) kW/hr per capita WDI

Labour (L) Labour force total WDI

Capital (K) Constant 2010 $ USD WDI

Ecological footprint (EFP) The global hectare of land GFP

15

appropriate technique when there exists a long-run equilibrium relationship among variables that are 209

integrated of I(1). The essence of VECM-Granger is to check the predictive power between the 210

variables to help craft effective policies.

211

3.2 Model Specification 212

The neoclassical aggregate production model proposed by Ghali and El-Sakka (2004) provides the 213

foundation for examining the relationship between electricity consumption and economic growth.

214

This model treats capital, labour and electricity (used as a proxy for energy) as separate inputs in the 215

production process. This model can be expressed as:

216

( , , , )

RGDP= f K L EU EFP (1)

217

To achieve homoscedasticity in the underlying data series, a logarithm transformation of equation (1) 218

is carried out.

219

𝑙𝑛𝑅𝐺𝐷𝑃 = 𝛿 + 𝛽₁𝑙𝑛𝐾 + 𝛽₂𝑙𝑛𝐿 + 𝛽₃𝑙𝑛𝐸𝑈 + 𝑙𝑛𝐸𝐹𝑃 + 𝜀_𝑡 (2) 220

A carbon-income function is formulated to investigate the trade-off between economic growth and 221

environmental degradation a phenomenon well known in the energy literature as the environmental 222

Kuznets curve (EKC) hypothesis (Shahbaz et al.,2013; Tiwari et al.,2013), presented as:

223

𝑙𝑛𝐸𝐹𝑃 = 𝛿 + 𝛽₁𝑙𝑛𝐾 + 𝛽₂𝑙𝑛𝐿 + 𝛽₃𝑙𝑛𝐸𝑈 + 𝛽₄𝑙𝑛𝐺𝐷𝑃 + 𝛽₅𝑙𝑛𝐺𝐷𝑃²+ 𝜀_𝑡 (3) 224

Where  represents constants and    ₁, ₂, ₃, ₄&₅ are partial slope parameters. K denotes capital, 225

this represents the capital stock in the production process; L denotes labour which represents the level 226

of employment in the production process; EC represents the total consumption of electricity, and 227

RGDP denotes real gross domestic product which represents the aggregate output of gross domestic 228

product. The constant parameter  and the partial slope coefficients  s, used in the model, measure 229

16

the marginal effect of capital and electricity on the output. In the production function earlier stated 230

posit long-run movement of variables may be connected (Ghali and El-Sakka 2004). In addition, to 231

account for the short-run dynamics in the factor-input behaviour, the functional specification in 232

equation (2) suggests that past behavioural changes in variables (capital, labour and electricity) can be 233

useful in predicting future changes of output (Lorde, Waithe and Francis, 2010). In a simple term, 234

causality can be used to investigate the relationship between the variables. The presents study draws 235

strength following the studies of Ghali and El-Sakka, (2004), Solarin (2011), Saidi and Hammami, 236

(2015), Shahbaz et al. (2016), Galli (2012), Dlamini et al. (2015), Mutascu (2016), Bimonte and Stabile 237

(2017), Sarwar et al. (2017), Amri, (2017), Destek, Ulucak, and Dogan (2018), and Akadiri et al. (2020).

238 239

3.3 Stationarity Test 240

Testing for stationarity among variables in time series analyses is required for establishing the order 241

of integration of the variables. This is essential for the avoidance of spurious regression. In 242

econometrics literature, several tests such as the Augmented Dickey-Fuller (1981), Phillips and Perron 243

(1988), and Elliot et al. (1992) tests can be applied to determine the order of integration of variables.

244

However, these conventional unit root tests are unable to account for the structural break(s) and are 245

thus prone to producing invalid and inconsistent estimates when structural break(s) exist in the data 246

series. Most macro-economic datasets are characterized by economic occurrences, which cause 247

structural breaks. Hence, this study balances with structural break unit root tests with Clemente, 248

Montanes and Reyes (1998) and Zivot-Andrews (1992) unit root tests which are known generally for 249

capturing structural breaks.

250

Zivot-Andrews test models are computed as stated below:

251

17

1 2 1

0 r

t t t i t i t

i

Y   t Y₋ DU Y₋ 

=

 = + + + +



252

1 2 1

0 r

t t t i t i t

i

Y   t Y₋ DT Y₋ 

=

 = + + + +



253

1 2 1

0 r

t t t t i t i t

i

Y   t Y₋ DU DT Y₋ 

=

 = + + + + +



254

There is a shift that occurs at each point of likely breaks at both intercept and trend or either one of 255

them as shown by the dummy variable DU. In the Zivot-Andrews unit root test, a null hypothesis of 256

unit root H₀: > 0 is tested against an alternative of stationarity H₁: < 0. This implies that failure 257

to reject H₀ indicates the presence of unit roots, while rejection confirms stationarity.

258

3.4 Procedures for Measuring Cointegration Relationships 259

There are numerous procedures documented in econometrics literature for testing cointegration 260

relationship among data series. The long-run relationship is said to exist between two series if there is 261

some sort of linear stationary combination among them (Engle & Granger, 1987; Johansen & Juselius, 262

1990; Phillips & Ouliaris, 1990; Johansen, 1991; Gregory & Hansen, 1996; Carrion-i-Silvestre & Sansó, 263

2006). However, all the above-mentioned cointegration tests render diverse conclusions of 264

cointegration and non-cointegration null hypotheses. More robust results can be obtained by exploring 265

the individual test statistics of Engle and Granger (1987), Johansen (1991), Boswijk (1995) and 266

Banerjee et al. (1998) as recently advanced by Bayer and Hanck (2013).

267

. .

2[log( _{rob EG}) ( _{rob JOH})]

EG−JOH = − P + P (7)

268 269

. . . .

2[log(( _{rob EG}) ( _{rob JOH}) ( _{rob BO}) ( _{rob BDM}))]

EG−JOH−BO−BDM = − P + P + P + P (8)

270

18

Where P_{rob EG.} ,P_{rob JOH.} ,P_{rob BO.} andP_{rob BDM.} are the individual probabilities of each of the test.

271 272

3.5 ARDL Approach 273

The ARDL bounds testing technique which guarantees more efficiency and robustness, especially in 274

small sample size, is used to test for cointegration among electricity consumption, economic output, 275

and ecological footprint (EFP). The merit of this technique is the possibility of both long and short- 276

run dynamics of the fitted regression with error correction model being reported at the same time as 277

well as determining the case of an unknown order of integration of series as long as the series is I(0) 278

and I(1), certainly not I(2). The unrestricted version of the error correction model is specified, and it 279

assumes that all variables are endogenous.

280

∆𝑌 = 𝛿₀+ 𝛿₁𝑡 + 𝛽₁𝑦_𝑡−1+ ∑^𝑧_𝑘=1𝛾₁𝑣_𝑘𝑡−1+ ∑^𝑋_𝑛=1𝜑_𝑛∆𝑌_𝑡−𝑛 + ∑^𝑍_𝑘=1∑^𝑋_𝑛=1𝜇_𝑘𝑛∆𝑉_{𝑘𝑡−𝑛}+ 281

𝜃𝐷_𝑡+ 𝜀_𝑡 (9)

282

𝐷_𝑡 is an exogenous variable which accommodates structural breaks in the framework, while Vk

283

represents the vector. F statistics computed from the bounds test is used to validate the null hypothesis 284

when there is no cointegration. Three different scenarios exist in making this decision: first, the 285

rejection of the null of no cointegration where the F-statistic computed is greater than the upper 286

bounds of the critical values reported. Second, an inconclusive cointegration where the F-statistic lies 287

within both lower and upper bounds. Third, a case of no cointegration where the F-statistic is below 288

the upper bound critical value. The specification of the hypotheses for bounds test is expressed as:

289

H₀:₁=₂ = =... _k+2 =0 (10) 290

H₁:₁ ₂  ... _k+2 0 (11) 291

19 3.6 Cointegration Estimation Techniques 292

The need to investigate the magnitude of long-run associations among variables is essential in time- 293

series estimation. The most widely known long-run estimators include the fully modified ordinary least 294

squares (FMOLS) advanced by Philips and Hansen (1990), the dynamic ordinary least squares (DOLS) 295

proposed by Stock and Watson (1993), and the Canonical Cointegration Regression of Park (1992).

296

These are useful methods that provide robust cointegrated regression estimates in cases where long- 297

run relationships exist. They are particularly efficient in small sample sizes.

298

3.6.1 FMOLS 299

The FMOLS method of cointegration estimation is distinct in its ability to provide optimal 300

cointegrating regression estimates among series integrated of order one (Phillips & Hansen, 1990;

301

Phillips, 1995; Pedroni, 2001a, 2001b). The approach also addresses the problem of endogeneity and 302

autocorrelation without compromising the robustness of the estimates.

303

𝑌_𝑖,𝑡 = ⍺_𝑖 + 𝛽_𝑖 𝑋_𝑖,𝑡+ 𝜀_𝑖,𝑡 ∀_𝑡= 1, … , 𝑇, 𝑖 = 1, … . . 𝑁 (12) 304

305

Allowing for 𝑌_𝑖,𝑡 and 𝑋_𝑖,𝑡 are cointegrated with slopes 𝛽_𝑖, where 𝛽_𝑖 may or may not be homogeneous 306

across i. Hence, the equation becomes:

307 308

𝑌_𝑖,𝑡 = ⍺_𝑖 + 𝛽_𝑖 𝑋_𝑖,𝑡+ ∑^𝐾_𝑘=−𝐾^𝑖 𝛾_𝑖,𝑘

𝑖 ∆𝑋_{𝑖,𝑡−𝑘}+ 𝜀_𝑖,𝑡 ∀𝑡 = 1,2, … , 𝑇, 𝑖 = 1, … . . 𝑁 (13) 309

310

We reflect 𝜉_𝑖,𝑡 = (𝜀̂_𝑖,𝑡, ∆𝑋_𝑖,𝑡) and 𝛺_𝑖,𝑡 = 𝑙𝑖𝑚

𝑇→∞𝐸 [¹

𝑇(∑^𝑇_𝑖=1𝜉_𝑖,𝑡)(∑^𝑇_𝑖=1𝜉_𝑖,𝑡)] as the long covariance. here 311

𝛺_𝑖 = 𝛺_𝑖⁰+ 𝛤_𝑖+𝛤_𝑖^´; The simultaneous covariance is depicted as 𝛺_𝑖⁰ while the weighted sum of 312

autocovariance is 𝛤_𝑖 . Thus, the equation of the FMOLS is rendered as:

313

20 314

𝛽̂_{𝐹𝑀𝑂𝐿𝑆}^∗ = ¹

𝑁∑^𝑁_𝑖=1[(∑^𝑇_𝑖=1(𝑋_𝑖,𝑡− 𝑋̅_𝑖)²)⁻¹(∑^𝑇_𝑖=1(𝑋_𝑖,𝑡− 𝑋̅_𝑖)𝑌_𝑖,𝑡^∗ − 𝑇_𝛾̂𝑖)] (14) 315

316

Where 317

𝑌_𝑖,𝑡^∗ = 𝑌_𝑖,𝑡^∗ − 𝑌̅_𝑖−^{𝛺̂2,1,𝑖}

𝛺̂_2,2,𝑖∆𝑋_𝑖,𝑡 𝑎𝑛𝑑 𝛾̂ = 𝛤̂_{𝑖 2,1,𝑖} + 𝛺̂_2,1,𝑖⁰ − ^{𝛺̂2,1,𝑖}

𝛺̂_2,2,𝑖(𝛤̂_2,2,𝑖+ 𝛺̂_2,2,𝑖⁰ ) (15) 318

319

3.6.2 DOLS 320

The DOLS technique is an alternative long-run equation estimator. It is known to possess merits over 321

FMOLS, and the unique feature of DOLS being efficient estimator asymptotically and also the ability 322

to eliminate feedback in the cointegrating system, DOLS can be substituted for FMOLS as advanced 323

by Saikkonen (1991) and Stock and Watson (1993). The estimation process of DOLS have lags and 324

leads in the cointegration regression.

325

𝑌_𝑡 = ⍺_𝑖 + 𝛽 𝑋´_𝑡+ 𝐷´_1𝑡𝐷´𝛾₁∑^𝑟_𝑗=−𝑞∆𝑋´_𝑡+𝑗⍴+ 𝑣_1,𝑡 (16) 326

From the above equation, the differenced explanatory variables with lag and lead of 𝑞 and 𝑟 327

accordingly absorb all the long-run relationship between 𝑣_1,𝑡 and 𝑣_2,𝑡 while the least-square estimates 328

of θ = (β', γ')' harbours asymptotic distribution parallel to CCR and FMOLS.

329

3.6.3 CCR 330

The OLS estimator has a shortfall when transforming variables in their second-order. Hence, the CCR 331

technique is exceptional in avoiding the bias of the second-order. The covariance matrix form of the 332

CCR is expressed as follows:

333

21 𝛺 = 𝑙𝑖𝑚_𝑛→∞ E ∑^𝑛_𝑡=1(𝑢_𝑡) ∑^𝑛_𝑡=1(𝑢_𝑡)^´=[𝛺₁₁ 𝛺₁₂

𝛺₂₁ 𝛺₂₂] (17)

334

From the above expression, Ω can be:

335

𝛺 = ∑ +𝛤 + 𝛤^´ (18)

336

and 337

∑ = 𝑙𝑖𝑚_𝑛→∞ E ∑^𝑛_𝑡=1(𝑢_𝑡𝑢´_𝑡) (19) 338

𝛤 = 𝑙𝑖𝑚_𝑛→¹

𝑛

E ∑^𝑛−1_𝑘=1∑^𝑛_𝑡=𝑘+1E(𝑢_𝑡𝑢´_𝑡−𝑘) (20)

339

⋂ = ∑ +𝛤 = (⋂_1,⋂₂ ) = [⋂₁₁ ⋂₁₂

⋂₂₁ ⋂₂₂] (21)

340

The series transformed obtained from above is given as:

341

𝑌_1𝑡^∗ = 𝑌_2𝑡− ∑ (⋂⁻¹ ₂ )´ 𝑢_𝑡 (22) 342

𝑌_2𝑡^∗ = 𝑌_2𝑡 − ∑ (⋂⁻¹ ₂ )´ 𝑢_𝑡 (23) 343

𝑌_1𝑡^∗ = 𝑌_1𝑡− ( ∑ (⋂⁻¹ ₂ 𝛽 + (0, 𝛺₁₂, 𝛺₂₂⁻¹ )´)´𝑢_𝑡 (24) 344

From the above, the long run estimator will acquire the following form:

345

𝑌_1𝑡^∗ = 𝛽´ + 𝑌_2𝑡^∗+𝑢_1𝑡^∗ (25)

346

From the outlined equation, the OLS estimators share the same style as the ML estimation. The 347

asymptotic endogeneity caused by the long-run correlation between 𝑦_1,𝑡 and 𝑦_2,𝑡 were avoided by the 348

transformation of the variables. The asymptotic bias due to cross-correlation between u1t and u2t is 349

resolved with the transformation of the variables expressed as:

350

𝑌_1𝑡^∗ = 𝑢_1𝑡− 𝛺₁₂, 𝛺₂₂⁻¹𝑢_2𝑡 (26)

351

22 3.7 Granger Causality Approach

352

Causality test is required to determine the direction of causality between variables as traditional 353

regression does not necessarily imply causal relationships. This is necessary to provide policymakers 354

and stakeholders clear insight into predictability powers that exist between variables. The expression 355

𝑋_𝑡 Granger causes 𝑌_𝑡 implies is that 𝑋_𝑡 (in its entirety i.e its present and past realizations) is a good 356

predictor of 𝑌_𝑡. Granger causality test in a bivariate form is specified as:

357

𝑋_𝑡 = 𝛿₀+ 𝛿₁𝑋_𝑡−1+ 𝛿₂𝑌_𝑡−1+ 𝜀_𝑡 (27) 358

𝑌_𝑡 = 𝛿₀+ 𝛿₁𝑌_𝑡−1+ 𝛿₂𝑋_𝑡−1+ 𝜀_𝑡 (28) 359

The null hypothesis that 𝑋_𝑡 does not Granger cause 𝑌_𝑡 is tested against the alternative hypothesis that 360

𝑋_𝑡 Granger causes 𝑌_𝑡. Granger causality relationships can take the following forms: (i) unidirectional 361

(implying either from 𝑋_𝑡 to 𝑌_𝑡 or otherwise), (ii) bidirectional (meaning feedback relationship from 𝑋_𝑡 362

to 𝑌_𝑡 and 𝑌_𝑡 to 𝑋_𝑡), and (iii) neutrality (this means there is no causal interaction between the variables 363

𝑋_𝑡 and 𝑌_𝑡).

364 365

3.7.1. The VECM Granger Causality Approach 366

The need for causality is crucial because of the directional causality flow and insight for policy and 367

decision-makers. The VECM approach is the most appropriate technique when there exists a long- 368

run equilibrium relationship among variables that are I(1). The Empirical construction of VECM 369

Granger causality is rendered as:

370

23

11 12 13 14 15

1

21 22 23 24 25

2

3 31 32 33 34 35

1

4 41 42 43 44 45

5 51 52 53 54 55

(1 ) (1 )

t i i i i i

t p i i i i i

t i i i i i

i

t i i i i i

t i i i i i

LnY L

LnK

L LnL L

LnEC LnEFP

    



    



     

     

     

=

     

     

     

     

− = + − 

     

     

      

   



1 1 1

1 2 2

1 3 1 3

1 4 4

5

1 5

t t

t t

t t t

t t

t t

nY LnK

LnL ECT

LnEU LnEFP









 

 

 

−

−

− −

−

−

     

     

     

 +  + 

     

     

      

   

(29) 371

Where (1−L) represents the difference operator, ECT_t−1 is lagged error correction term. _it is the 372

stochastic term (disturbance term) which is required to be IID~N(0, ) meaning that disturbance term 373

is independently identically normally distributed with constant variance and zero mean. T-statistic 374

indicate a long-run causal relationship between the variables.

375

376

4. Results and Discussion 377

A graphical representation showing the behaviour of the dataset used in the time series estimations is 378

depicted in Figure 2. The possibility of a structural break is evident in Figure 2, informing our decision 379

to account for structural breaks in the estimation process. The descriptive statistics that renders the 380

basic summary statistics like mean, median, standard deviation, data distribution (reported by Kurtosis 381

and Jargue Bera) and correlation coefficients matrix are presented in Table 3. The Jarque Bera test 382

statistic in Table 3 reports that all the variables are normally distributed (p-value >0.05). Though there 383

is a huge difference between the minimum and maximum values for the period investigated. This 384

suggests a need for further tests. The correlation analysis reports a positive and statistically significant 385

relationship between electricity consumption and the economic output (GDP). The ecological 386

footprint has a positive interaction with economic growth. The association established between the 387

variables cannot be statistically inferred, hence, requires subsequent econometric estimation for 388

statistical inferences.

389

24 390

Figure 2: Graphical representation of RGDP, EC and EFP in logarithm form 391

392 393 394

25

Table 3: Descriptive Statistics and Correlation Analysis

lnEC lnEFP lnK lnL lnRGDP

Mean 7.453377 1.055078 25.64037 16.92926 9.091968

Median 7.419034 1.036616 25.52474 16.90245 9.017334

Maximum 7.956675 1.223487 26.35993 17.17263 9.496455

Minimum 6.834862 0.84991 24.9895 16.77223 8.81122

Std. Dev. 0.353451 0.110373 0.448173 0.10668 0.209281

Skewness -0.18471 -0.20913 0.139954 0.848321 0.416491

Kurtosis 1.842195 2.067187 1.627793 2.895078 1.977383

Jarque-Bera 1.538529 1.088619 2.043021 3.010006 1.812087

Probability 0.463354 0.580242 0.360051 0.222017 0.40412

Sum 186.3344 26.37695 641.0093 423.2314 227.2992

Sum Sq. Dev. 2.998264 0.292373 4.820608 0.273135 1.051169 Correlation Matrix Analysis

lnEC lnEFP lnK lnL lnRGDP

lnEC 1.0000

t-Stat -

Prob -

lnEFP 0.8620^*** 1.0000

t-Stat 8.1555 -

Prob 0.0000 -

lnK 0.9436^*** 0.9464^*** 1.0000

t-Stat 13.6738 14.0525 -

26

Prob 0.0000 0.0000 -

lnL 0.9000^*** 0.7657^*** 0.8506^*** 1.0000

t-Stat 9.9023 5.7103 7.7602 -

Prob 0.0000 0.0000 0.0000 -

lnRGDP 0.9614^*** 0.9067^*** 0.9803^*** 0.9299^*** 1.0000

t-Stat 16.7740 10.3099 23.8128 12.1323 -

Prob 0.0000 0.0000 0.0000 0.0000 -

Source: computation by Authors

395 Note: ***, ** and * indicate 1%, 5% and 10% statistical significance level respectively

396 397

This study proceeds to investigate the stationarity properties of the investigated variables using a 398

battery of unit root and stationarity test. This is necessary to ascertain the accuracy of the estimates, 399

thereby providing the needful policy insights. The results of the stationary/unit root test are reported 400

in Tables 4 and 5. Precisely the ADF and PP, results are in harmony of variables integrated of order 401

one. Although, the ERS unit root test renders mixed results. Thus, the need to investigate the variables 402

using the KPSS stationarity test. The KPSS with reverse null hypothesis supports the integration of 403

order 1. The consensus of the results declares that the variables are integrated of order one, ~I(1).

404

Subsequently, the Zivot and Andrews (1992) and the Clemente-Montanes-Reyes-structural break 405

detrend unit root test results with simple structural break dates are reported in Table 5. The results of 406

the break test of ZA and Clemente-Montanes-Reyes-structural break detrend unit root test results 407

corroborate the integration status of the variables. These identified break dates correspond with 408

significant economic and political events in Turkish history.

409

Table 4: Unit Root Tests 410

Variables ADF PP ERS DF-GLS KPSS ZA

27

lnEC -1.8263 -1.7198 15.3736^*** -2.8079 2.1308^** -3.6691 (1) [2001]

∆lnEC -4.2171^*** -5.0137^*** 3.4264 -4.4515^*** 3.1399 -4.9266^* (1) [2004]

lnRGDP -2.0424 -2.1196 13.9451^*** -2.1705 2.1457^** -3.5459 (1) [2001]

∆lnRGDP -4.8769^*** -4.8766^*** 7.4965^*** -5.0918^*** 0.0464 -5.1214^** (1) [2003]

lnEFP -2.6698 -1.6979 7.5376^*** -4.7507^*** 3.0867^** -5.8043^*(1) [2001]

∆lnEFP -4.6537^*** -10.2486^*** 11.3365^*** -8.7275^*** 0.0995 -9.1528^***(2) [2003]

lnK -3.3665 -3.3605^* 8.3731^*** -3.4625^** 4.0832^*** -4.4499 (1) [2003]

∆lnK -6.7221^*** -6.7671^*** 8.9450^*** -6.9434^*** 0.0780 -7.2603^**(1) [2003]

lnL -0.6452 -0.3619 25.6038^*** -1.0496 3.1513^** -3.8856 (1) [2001]

∆lnL -5.7006^*** -5.7006^*** 8.0736^*** -5.8887^*** 0.1138 -7.0600^** (1) [2000]

Note: ***, ** and * indicate 1%, 5% and 10% statistical significance level respectively. []break year while () denotes optimal lag length. All tests are

411

conducted with a model of both intercept and trend orientation.

412 413

Table 5: Unit root with structural break using Clemente-Montanes-Reyes Test 414

Variables Innovative outliers^†

break^† Additive Outlier^†

Break^†

lnEC -0.151 2002 -2.216 2004

∆lnEC -4.27^** 2000 -5.347^** 1999

lnRGDP -1.541 2002 -2.151 2007

∆lnRGDP -5.25^** 2000 -4.33^** 1999

lnEFP -4.508 2004 -4.769 2003

∆lnEFP -9.239^** 2000 -6.199^** 1999

lnK -3.139 2002 -3.518 2003

∆lnK -7.283^** 2000 -4.805^** 1999

lnL -1.469 2007 -2.382 2009

28

∆lnL -4.484^** 2007 -7.053^** 2007 Source: Authors computation from STATA 15.0 software

415 Note: ***, ** and * indicate 1%, 5% and 10% statistical significance level respectively

416 417 418 419 420 421

Table 6: Lag criteria selection or maximum lag length selection 422

Lag LogL LR FPE AIC SC HQ

0 159.4791 NA 1.77E-12 -12.87326 -12.62783 -12.80814

1 271.8332 168.5312* 1.28e-15* -20.15277* -18.68020* -19.76210*

Source: Authors computation from E-views 10.0 software

423 Note: LR denotes sequential modified LR statistic, FPE represents Final prediction error. AIC stands for Akaike information criterion,

424

while SIC means Schwarz information criterion and finally Hannan Quinn information for HQ.

425 426 427

The maximum lag length selection criteria are presented in Table 6. These selection criteria offer the 428

opportunity for a parsimonious model to be chosen. From Table 6, the most appropriate criteria for 429

selection is Akaike Information Criteria (AIC) which can accommodate sample size and suitable for 430

the nature and structure of this study (Lutkepohl, 2006).

431

The next step is the establishment of long-run equilibrium relationship (cointegration) via a battery of 432

cointegration techniques namely Bayer & Hanck (2013) combined cointegration in conjunction with, 433

Pesaran ARDL bounds test and Maki (2012) cointegration test. All aforementioned cointegration tests 434

are in the consensus of a cointegration relationship between electricity consumption, economic growth 435

29

ecological footprint, capital and labour over the investigated period. This implies that there is some 436

sort of convergence among the variables. The use of Maki cointegration test is to capture the possible 437

structural break given the robustness of the test to accommodate up to 5 structural breaks⁴. 438

The Bayer & Hanck cointegration test results are reported in Table 7, confirming the presence of an 439

equilibrium relationship among the series investigated (p-value < 0.01). Thus inferring a long-run bond 440

between the outlined variables. For precision and robustness check, an ARDL bounds test is 441

conducted to validate the results of the Bayer and Hanck as documented in the appendix section.

442

Table 7: Bayer and Hanck result

Fitted Model EG-JOH EG-JOH-BO-BDM Cointegration Remark

lnRGDP= f(lnk, lnL, lnEC, lnEFP) 70.464^*** 180.988 Yes

lnEFP= f(lnGDP, lnGDP², lnEC, lnK, lnL) 56.624^*** 167.148 Yes

Source: Authors’ Computation.

443

***, ** and * denote 1%, 5% and 10% statistical significance level respectively

444 445 446

Table 8: ARDL long-run and short-run results

Model RGDP = f(lnK, lnL, lnEC, lnEFP) LNEFP= f(ln^K, ln^L, ln^EC, ln^RGDP, ln^RGDP2) Variable Coefficient Std error t-stat Coefficient Std error t-stat

Short-run results

ECT(-1) -0.7275^* 0.3284 -2.2151 -0.7052^* 0.1291 -5.4638

∆lnK 0.4245^* 0.0964 4.4025 0.3499^*** 0.1893 1.8482

∆lnL 0.4031^* 0.1052 3.8298 0.6035^* 0.2776 2.1737

∆lnEC 0.3898^** 0.1457 2.6746 0.3449^** 0.1561 2.2088

4 More details regarding Maki cointegration test can be provided upon request. Although the test is reported in the appendix section. The results is in harmony as ARDL bounds test and the Bayer and Hanck cointegration results

30

∆lnEFP -0.0659^*** 0.0306 -2.1485

∆lnRGDPC 0.7144^** 0.3357 2.1284

∆lnRGDPC² -0.8229^** 0.3723 -2.2102

Constant -17.8533^* 3.7392 -4.7746 11.1077^* 4.4874 -2.4743 Long-run results

lnK 0.4191^* 0.1386 3.0238 0.3466^** 0.1732 2.0013

lnL 0.9928^* 0.2093 4.7434 0.5978^** 0.2964 2.0171

lnEC -0.0651^** 0.0273 -2.3806 0.3416^** 0.1671 2.0442

lnEFP -0.3341^*** 0.1781 -1.8767

lnRGDPC 0.8376^** 0.4005 2.0916

lnRGDPC² -0.9132^** 0.4229 -2.1425

Constant -17.6247^* 2.3077 -7.6373 -11.5773^** 4.9669 -2.3309

Source: Authors’ computation

447

*, ** and *** denote 1%, 5% and 10% statistical significance level respectively

448 449

Table 8 presents the ARDL long and short-run results which affirm the long-run equilibrium 450

relationship for all the estimated models. This implies that there is convergence among the variables 451

(RGDP, EFP, K, L and EC). The validation of the long-run relationship is evident in the rejection of 452

the null hypothesis. Table 8 reveals a very high speed of adjustment of over 70% with the contribution 453

of the regressors. Both capital and labour contribute to economic growth and environmental 454

degradation in both short and long-run. More precisely, a 1% increase in K stimulates GDP and EFP 455

at ~0.34% and ~0.41%, respectively both in short- and long-run. This outcome is indicative of 456

policymakers, as capital and labour accumulation are the key drivers of growth in Turkey. This finding 457

is in line with the Solow Growth Model and Soytas and Sari (2009). Energy (electricity) consumption 458

31

increases environmental degradation and economic growth, meaning that, Turkey’s economy is 459

energy-dependent. A 1% increase in EC stimulates EFP at ~0.34% both in short- and long-run, 460

whereas GDP at 0.38% increase and 0.06% decrease in short- and long-run, respectively. These results 461

corroborate with others in the literature such as Farhani and Ozturk (2015); Al-Mulali et al. (2015a, 462

b). This is in line with the electricity-led growth hypothesis, thus, caution is advised in the adoption of 463

conservative energy policy measures in order not to jeopardize economic growth. As such, any action 464

on the path to apply energy cut will harm economic growth. This is consistent with the study 465

conducted for Zimbabwe by Samu et al (2019). However, energy (electricity) consumption in the long- 466

run has a negative statistical impact (P<0.10) on economic growth. This is insightful for decision- 467

makers that in the long-run intensification of energy will harm economic growth. This is further 468

reinforced by the outcome of environmental degradation on economic growth. We observe a trade- 469

off between economic growth and environmental quality. This phenomenon re-echoes the 470

Environmental Kuznets Curve (EKC) hypothesis. This indicates that Turkey’s economy is yet to attain 471

its environmental target. This implies that a scale stage development as an emerging economy where 472

economic growth has priority over environmental quality (Shahbaz & Sinha, 2019).

473

The fitted model in Table 8 further affirms the significant contribution of capital and labour stock to 474

economic output in both the long and short run. The striking revelation of the model is the affirmation 475

of the EKC hypothesis for Turkey both in the short-run and in the long-run. This is consistent, as a 476

statistical positive sign for GDP and negative sign of squared GDP are observed. This implies an 477

inverted U-shaped characteristic in the relationship between economic output and environmental 478

quality. This unique shape explains that the environmental quality declines first as economic growth 479

increases until a certain threshold of GDP, where environmental quality increases with increasing 480

economic output (Saboori et al. 2012; Fodha and Zaghdoud, 2010). From the initial economic growth 481

stage (scale stage) there is little or no environmental consciousness in the course of increasing 482